[1O5-OS-18b] OS-18

[1O5-OS-18b-03] Development of a convolutional neural network model for prediction of Alzheimer's disease indicators

〇Takumi Sekiya¹, Kazutaka Sakamoto², Ryota Kobayashi², Shinobu Kawakatsu³, Keigo Matsuda¹ (1. Nagoya University, 2. Yamagata university, 3. Fukushima Medical University)

Keywords:Alzheimer’s Disease, MRI, Convolutional Neural Network

The number of patients with AD is increasing worldwide. In order to stop the progression of AD before the onset of dementia, it is necessary to detect the disease at an early stage and provide treatment to control the progression. Until now, MMSE scores and MRI images have been used for diagnosis, but it is difficult to fully utilize them as data for early detection of AD because the changes in brain structure are small in the early stage of AD. In this study, we constructed a model that detects the features of brain structure from MRI images by machine learning and predicts three indices: severity, which indicates the degree of blood flow reduction, GM atrophy, which indicates the degree of atrophy of GM where neurons gather, and MMSE, and evaluated the model performance. The PCC of the regression model with MMSE, severity, and GM atrophy as outputs was found to be 0.997, 0.933, and 0.992, respectively.